fix(scheduler): detect auth-failure markers in session output

[brianjones-v4n]

Summary

Claude Code exits cleanly (status completed, exit code 0) on a 401 / invalid-auth response. Without scanning the captured session output, the scheduler records these silent auth failures as successes — the job appears healthy while authentication is actually broken.

This adds a post-completion scan of the last 4 KB of session output against well-known auth-failure regex landmarks:

• 401 Unauthorized / 401 ... authentication
• Invalid (authentication|API key|bearer token|x-api-key)
• Unauthorized
• OAuth token has expired / is invalid / expired / missing
• API key not valid / found / provided
• Please (re-)?authenticate

On hit:

• Override failed = true
• Surface the matched fragment as the error reason in runHistory.recordCompletion's error field
• Same reason flows into the persisted lastError for the job

Test plan

• Test asserts the AUTH_FAILURE_PATTERNS const exists at module scope with expected regex landmarks
• Test asserts the completion handler uses let failed so the override is possible
• Test asserts the scan reads only the last 4 KB of output
• Test asserts the auth-failure reason flows into both error and lastError
• npx tsc --noEmit — no new errors (9 pre-existing errors unchanged)

🤖 Generated with Claude Code

[vercel]

@brianjones-v4n is attempting to deploy a commit to the sagemind Team on Vercel.

A member of the Team first needs to authorize it.